Most companies arrive at a multilingual website along the same path. A first order comes in from Greece or Romania, then a second one, then somebody asks whether there is an English version. The fix looks simple: install a plugin, press a button, done. Three months later it turns out Google is indexing only the original language, the English pages are competing with each other, and the German translation contains sentences that make no literal sense.
Translating a website is an engineering project, not a copy task. This article walks through the decisions that have to be made before the first translated word, and through the mistakes that cost the most.
When a multilingual site actually makes sense
Before the technical work comes a sober assessment. A second language doubles the volume of content that has to be maintained. Every new product page, every change to shipping terms, every blog post gets multiplied by the number of languages.
It makes sense when you already have proven demand from abroad. Check the country report in Google Analytics 4 and the queries by country in Search Console. If 15 percent of traffic comes from Romania and you have not a single line in Romanian, that is a signal. Calls or emails in English are an even stronger one.
It also makes sense when the product is logistically ready for export. A multilingual site without clear international shipping terms and without prices in the right currency simply moves the problem one step back.
It does not make sense when the English version is built for prestige and nobody will maintain it. An abandoned, half empty section does more harm than no section at all.
Which language first
English is the sensible default for B2B and services, because it covers an international audience without a geographic boundary. For B2C and physical goods it is often smarter to start with a specific neighbouring market: Greek, Romanian, Serbian. Search in the local language carries lower competition and higher purchase intent than generic English.
Three structural approaches to URLs
This decision is made once and is hard to reverse, so it deserves attention.
| Approach | Example | Upside | Downside |
|---|---|---|---|
| Subfolder | example.com/en/ | Inherits domain authority, easy to maintain, cheapest | Weaker local signal for a specific country |
| Subdomain | en.example.com | Clear separation, separate hosting if needed | Authority accumulates separately, more technical work |
| Separate domain | example.ro | Strongest local signal, trust on the local market | Each domain is promoted separately, highest cost |
For the large majority of companies the subfolder is the right choice. It preserves the accumulated authority of the main domain, requires no new infrastructure and is the simplest to maintain. Separate domains are justified only with a serious presence on a specific market, a local team and a dedicated marketing budget.
What sits inside the URL matters too. Paths get translated: /en/services/ rather than /en/uslugi/. Transliterated source language words inside a foreign URL confuse both the visitor and the search engine.
hreflang: the signal most sites get wrong
hreflang tells Google which version of a page to show to which user. It looks simple, yet it is among the most frequently misimplemented pieces of technical SEO.
Three rules solve almost every problem.
First, the references must be reciprocal. If the Bulgarian page points to the English one, the English page must point back. A one way declaration is ignored.
Second, every page references itself. The hreflang set must include a line for the page it sits on.
Third, use absolute URLs and canonical versions. If a page has a canonical pointing elsewhere, hreflang must point at the canonical, not at a parameterised variant.
Add x-default for visitors whose language you do not cover. Usually that is the English version or a language selection page.
Codes follow the standard: en for English generally, en-GB for the United Kingdom, de-AT for German in Austria. A combination like en-EU does not exist and is ignored.
Machine translation versus edited translation
The quality of machine translation has changed radically in recent years. DeepL and modern language models produce output that works as a draft. The problem is not grammar but terminology and context.
A sensible workflow has three layers. The machine produces the first version. A person who knows the industry edits terminology and tone. Finally, the elements machines systematically get wrong are checked: product names, units of measure, legal texts, calls to action.
Some categories of content should never ship as unedited machine output. Terms and conditions and privacy policies carry legal risk. So do product descriptions in regulated sectors such as food, cosmetics or medical devices. Titles and meta descriptions deserve manual work, because they are the first thing a user sees in search results.
For blog content and help articles, edited machine translation is usually enough and cuts cost substantially. If you publish regularly, see our approach to copywriting.
What gets translated besides the visible text
This is where most gaps hide, because these elements are invisible during a casual page review.
Meta title and meta description on every page. Left in the source language, the translated version looks broken in search results.
Image alt text. It carries both accessibility and indexing value.
Button labels, form error messages, post submission confirmations, automated emails. A customer who fills in an English form and receives a confirmation in another script loses trust instantly.
Text baked into images. A banner with copy inside the graphic needs a separate file per language. That is why it is better practice to place text as HTML over the image rather than inside it.
Structured data. If you have schema markup with descriptions and prices, it gets localised too. There is more on this in our article on structured data.
Date, number and currency formats. 1,500.00 EUR and 1 500,00 EUR are not the same thing to a reader, and the wrong format reads as a technical fault.
Multilingual online store: where it gets harder
In a store, commercial decisions stack on top of the language layer, and they are not translation problems.
Currency does not follow language. A Greek speaker in Cyprus and one in Greece are both in the eurozone, but an English speaker in the UK expects pounds. Currency selection has to be independent from language selection and remembered between sessions.
Tax treatment differs by country for consumer sales inside the EU. Thresholds and the OSS regime require configuration in the system, not just a note on the page.
Shipping is the most common blocker. If the calculator shows no price and no delivery window for the specific country, the cart gets abandoned. Terms should be localised and shown on the product page, not only in an information section.
Customer reviews and questions usually stay in their original language. That is acceptable and even builds trust, as long as the interface around them is translated.
For practical launch steps see our online store checklist and our online stores service.
Common mistakes that cost traffic
Automatic redirection by IP address. A local user looking for the English version is forcibly sent back and cannot escape. The Google crawler is redirected too and never indexes the other versions. The correct approach is a suggestion banner, not a forced redirect.
Mixed languages on one page. English menu, source language content, half translated footer. This happens when translation is done piecemeal.
No translation of the 404 page and the search results page. A visitor who lands there falls out of the language context.
Duplicate content between en and en-US versions that are identical. If there is no real difference, maintain one English variant.
A language switcher hidden in the footer as flags. A flag denotes a country, not a language, and the Swiss flag maps to no single language. Written language names work better.
How to plan the project
A realistic sequence looks like this. First choose the URL structure and lock it. Then audit the content and decide which pages launch in the second language, because it is rarely necessary to do everything at once. Usually you start with the home page, services or top products, contact and legal texts.
Next comes the technical implementation: hreflang, a sitemap with language variants, the switcher, localised formats. Then translation and editing. Finally, set up separate tracking in GA4 so you can see how each language segment behaves. For the setup see our guide to Google Analytics 4.
Cost depends on the number of pages, the platform and the level of translation editing. WEBPROGRESS quotes each project individually and replies within 24 hours of receiving the brief.
Frequently asked questions
Plugin or custom implementation
On WordPress, plugins are a perfectly acceptable solution and save considerable time. What matters is choosing one that creates real separate URLs and correct hreflang, rather than one that translates dynamically with a browser script. Dynamic translation is not indexed and brings no organic traffic.
How many languages is reasonable
As many as you can keep current. Two well maintained versions deliver more than six abandoned ones. The practical rule is to add a language only when there is a measurable demand signal and someone to update it at least monthly.
Does being multilingual slow the site down
Not in itself. Specific implementations do: heavy plugins that load dictionaries on every request, or duplicated assets per language. With sound architecture the difference is negligible. For measurement see PageSpeed Insights.
Do I need a separate domain per market
In most cases no. A separate domain is justified by a serious long term investment in a specific country, with a local team and a dedicated budget for advertising and links. For a company testing a new market, the subfolder is faster and cheaper, and moving to a separate domain later is always possible.
How do I measure whether the second language is worth it
Track organic traffic by language folder, conversion rate per segment and revenue per visitor. Give the new version at least six months, because indexing and ranking take time. If after a year the English version delivers under 5 percent of enquiries, revisit the language choice.
Conclusion
A multilingual site works when the technical foundation is right from day one: clean URL structure, correct hreflang, localised formats and translation that has passed through a human. Everything else is an add on.
If you are considering a new market, or you already have foreign traffic that does not convert, write to us through contact. We will review the site, propose a structure and send an individual quote within 24 hours. See also our services for web development, SEO and GEO and advertising.
